Investigation of G007-LK treatment in mouse and human melanoma cell lines

zero_replacement <- function(zeroes_removed_df, cols_to_join){
# Replace the zero values in the dataset with a calculated minimum-non-zero value per observation multiplied by 0.65. Thus we ensure that the 0 values that cannot undergo log transformation are replaced by a small non-zero value.
zeroes_replaced_df <- zeroes_removed_df |>
  rowwise() |>
  mutate(min_non_zero = min(c_across
                            (where
                              (is.numeric))[c_across(where
                                                     (is.numeric)) != 0], na.rm = TRUE)) |>
  ungroup() |>
  mutate(across
         (where
           (is.numeric), ~ ifelse(. == 0, 
                                  0.65 * min_non_zero,
                                  .))) |>
  select(-min_non_zero)

#This part calculated the sum across all genes for every observation in the dataset with the replaced zeroes
zeroes_replaced_sums_df <- zeroes_replaced_df |>
  rowwise() |>
  mutate(total_sum = sum(c_across
                         (everything()),
                         na.rm = TRUE))

n = 1000000 # the data are normalized in TPM that is why we selected 1 million as n

#Data normalization. This part ensures that different experiments with varying total expression levels are brought to consistant scale
scaled_data_df <- zeroes_replaced_sums_df |>
  mutate(scaling_factor = n / total_sum) |>
  rowwise() |>
  mutate(across(-c(total_sum, scaling_factor), ~ . * scaling_factor)) |>
  select(-total_sum, -scaling_factor)

scaled_data_df |>
  rowwise() |>
  mutate(row_sums = sum(c_across
                        (everything()), 
                        na.rm = TRUE))

#This part of the code binds the normalized and scaled expression data with the first columns from the original df data.
joined_df <- bind_cols(cols_to_join,
                          scaled_data_df)
print(joined_df)
}